I am in a similar situation with firmware 2.0RTORC4.
I am using Sensorless Homing (stall effect) for G30 to find the Z=0. Then I would use G29 with a capacitive probe for mesh leveling. I found that I can run G30 and then manually move to Z0 with great precision, then I run a G30 S-1 with the capacitive probe and I get the offset value. I would use this offset for mesh leveling, but when it comes to the print, the height would be off. I have narrowed down the issue to temperature drift of the probe which, added to the natural tolerance error of the sensor, can vary a lot from a cold bed to a hot bed.
Now I am in the process to test if I can compensate this temperature drift with G31 using the S and C parameters that would allow to define some reference temperature and linear compensation, but I need to do some measurements to calculate the C coefficient before I put this to work and see repeatability patterns and improvements.
....also my capacitive probe is mounted in parallel to the ED3 J6 and receives the air flow from the cooling fan, which throws hot air and would create a micro climate around the probe, which might not be so good to fight thermal drifts....
